The implementation of the deep learning approach in Indonesia's national curriculum requires vocational schools to design instructional media that support mindful, meaningful, and joyful learning experiences, particularly for taxation-related subjects that students often find difficult to grasp. This research was conducted to develop a Construct 3-based learning media for the Taxation subject for Grade XI students of the Accounting and Financial Institution Program at State Vocational High Schools in Surabaya. The study was motivated by the findings of a preliminary investigation, which showed that classroom instruction was predominantly conducted through lecture methods and supported by conventional learning media, such as PowerPoint slides, textbooks, and student worksheets. Consequently, students tended to exhibit low learning engagement and experienced difficulties in understanding the subject matter. The questionnaire results further revealed that 72.3% of the respondents encountered challenges in comprehending Value Added Tax (VAT) and Luxury Goods Sales Tax (LGST) topics. This study employed a Research and Development (R&D) approach using the ADDIE model, consisting of Analysis, Design, Development, Implementation, and Evaluation stages. The resulting product is an interactive learning application equipped with several main features, namely Learning Outcomes and Learning Objectives Flow (CP-ATP), learning materials, Tax Insight, a tax calculator, practice exercises, and a developer progile. The results of the material expert validation indicated a feasibility percentage of 93%, categorized as highly feasible, while the media expert validation achieved a perfect score of 100%, also classified as highly feasible. A limited field trial was carried out involving 20 Grade XI Accounting students from four State Vocational High Schools in Surabaya. The students' responses to the developed learning media yielded a percentage of 98%, reflecting a strong level of agreement regarding its use in the learning process. These findings demonstrate that the Construct 3-based learning media incorporating a deep learning approach is highly feasible and effective in supporting technology-enhanced learning in the Society 5.0 era.